生成式AI難以終結SaaS，但將淘汰劣質監管科技

U.S. banks have long used rules-based engines and predictive models to screen transactions, but high false-positive rates and labor-intensive investigations are pushing them toward generative AI and large language models. The tools can summarize cases, search customer files and draft suspicious activity narratives. Their use matters because obligations under the Bank Secrecy Act remain with the institution: faster workflows do not remove the need for explainability, audit trails, data controls and accountable human judgment.

FinCEN’s April 10, 2026 proposal to overhaul AML/CFT program rules explicitly cited machine learning and GenAI as tools institutions could evaluate. A week later, on April 17, the OCC, Federal Reserve and FDIC issued SR 26-2, replacing older model-risk guidance while excluding generative and agentic AI from its scope. Regulators remain concerned that LLMs can produce persuasive but factually wrong output, a risk unlike errors in conventional predictive models. No transaction value, investment amount or penalty was disclosed in connection with the development.